ction 
form 
ts the 
cate a 
ed by 
eis a 
facts 
idded 
ecific 
n tree 
lated 
to the 
» is tO 
naller 
ships 
ulting 
ilting 
main 
direct 
s and 
ss by 
e first 
n of a 
iodels 
a), by 
ch as, 
nation 
aryıng 
i then 
  
[ INTELLIGENT INTERFACE 
INTERFACE LAYER 
SLEMS DBMS INTERFACE 
Nu 
  
à 
  
  
  
  
LEARN # THE CDATA s] 
SUB SYSTEM 
EXPERT FUZZ INTERFACE SCHEMA 
I ^ THE CDATA 
DBMS 
  
C INFERENCE LAYER Jj 
Y: 
(Low LEVEL OPERATORS ) 
iis: 
SLEMS 
KNOWLEDGE BASE 
  
  
  
  
    
    
À 
  
  
  
  
  
  
SLEMS 
DATABASE 
  
Figure 3: The SLEMS System Configuration (source, E.G. Mtalo, 
1990, page 72) 
the value of the runoff curve number (CN) for the particular area 
must first be determined. The CN is an empirical measure of the 
runoff potential of a region which together with two parameters 
S', P'(functions of the basin retention potential and 
precipitation) is used to calculate the potential runoff (Q) and 
soil loss A(Q, S,L,K,C,P)(Bondelid, T. R. et al., 1980). If the 
CN values are not available default values must be estimated. 
This, in turn, requires the choice of a specific method such as 
the US Department of Agriculture method (USDA AMC-II CN 
METHOD), the US Geological Surveys method (USGS LUDA 
CN METHOD) or from Landsat data (LANDSAT CN 
METHOD) as indicated in Figure 4. 
When a specific model has been chosen parameter values must 
be determined. Missing parameters must be computed or 
adopted from default values. This in turn requires further choice 
of a parameter estimation method(USLE PRMS in Figure 4) 
which depends on the required parameter and USLE model 
chosen. If, for example, the value of the slope factor S is 
missing and the terrain slope is known (e.g. from topographic 
maps or digital terrain models) different formulae must be used 
to estimate its value depending on the magnitude of the slope 
gradient and the length of the slope(Figure 4). The estimated 
value is then plugged into the appropriate USLE model (e.g. 
A(R,S,K,L,C,P)) to compute the soil loss. 
Figure 5 illustrates how specific knowledge on the selection of 
default C factors for the Universal Soil Loss Model(USLE) can 
be organised by method and location. Methods can be classified 
by inventor(e.g. HOLY and WISCHMEIER) or proprietor(e.g 
USDA). Within each subcategory (e.g. HOLY) default 
parameters may be selected according to the crop growth period 
(PERIOD 1, PERIOD 2 ... etc). Where location specific 
parameters are available (e.g GRANDFALLS) default C factors 
may be organised according to crop type (e.g. POTATO, 
GRAIN and HAY or POTATO and BROCCOLI). The selection 
of default parameters within each crop type may then be done 
according to the Crop Rotation Cycle such as P/P/P/G/H 
indicating three potato seasons followed by grain and hay 
rotations(Figure 5). 
The solution of soil erosion problems requires inputs from a 
wide range of sources. In particular knowledge about ground 
cover (e.g. crop type and crop rotation cycle) is essential in 
201 
estimating the C factor, which is, a measure of the degree of 
protection offered by ground cover against rain (and wind) 
erosion. Plant cover parameters, such as cover type and density, 
can be assessed from aerial photographs by aerial 
photointerpretation or by digital image analysis techniques. The 
knowledge needed to extract such information can be analyzed 
and organised into a semantic network of related of knowledge. 
Figure 6 shows the decision tree resulting from the application 
of this strategy to the method of dichotomous 
photointerpretation of crop cover from aerial photographs. In 
this case each node in the tree represents a binary decision 
where the photo interpreter must make a choice between two 
alternatives as indicated on the associated key(Figure 6). The 
terminal nodes indicate the required classes. SLEMS can 
therefore be used to assist in the visual interpretation of ground 
cover based on the stored aerial photointerpretation knowledge. 
Using a similar strategy the knowledge needed to facilitate soil 
taxonomic classification can be analyzed and represented by a 
semantic network. 
In order to compile knowledge on a particular subject, the 
domain knowledge must first be analyzed and organised as 
explained. The resulting decision tree must then be translated 
into "IF... THEN..." rules which are then compiled and stored 
by the system's rule editor. Using the SLEMS EXPERT a rule 
base containing: rules for the selection of soil erosion models 
and their parameters, soil taxonomic classification rules, rules 
for the selection of spectral bands and their application to the 
classification of ground features, and the procedure for 
dichotomous air-photo interpretation was successfully 
compiled. 
The system can also be used to record knowledge on the 
systems internal organization. The rules resulting from this are 
referred to as meta rules and they contain knowledge about the 
system and its application. Meta rules provide guidance to the 
user in the application of the knowledge contained in the 
knowledge base. They can also provide guidance in the 
compilation of knowledge. 
A demo prototype of SLEMS was introduced to a group of 
multi-disciplinary local experts at the Ardhi Institute in Tanzania 
who were then asked to comment on the viability of the expert